you write in the book (last row, p.32). "Table 2.3 shows the obtained results for different subsets of words present in the e-mail". This phrase is misleading because it is not very precise with respect to the role of the words not contained in the subsets. These could take the role of ’Free’ variables or of ’Known’ but not observed variables. Under the first interpretation ("partial instantiation") you have for Table 2.3 one to three observations, under the second ("full instantiation") you have for all 5 rows 5 observations. I reconstructed the spam filter with WebCHURCH and found out that the posterior P’s are computed under "full instantiation", so that the ’Free’-set is always the empty set.
You come to the same conclusion, when you study the ProBT-code.
I have been trying to create models, by using functions described on file:///C:/Users/Antoine/Desktop/ProBtWindows/Documentation/annotated.html .
Some of them work fine, but I have faced difficulty implementing the functions which require input parameters which types were defined with the associated object.
For example, when I try to use the function "Generate_Sequence", from HMM (file:///C:/Users/Antoine/Desktop/ProBtWindows/Documentation/d6/d2c/a00072.html), I need to use an "Observation_Sequence_t" object as an input. The "Observation_Sequence_t" type was defined as a public type related to the Object "HMM", but it does not seem to appear in the "pypl" library I have downloaded. Hence, when I try to define an "Observation_Sequence_t" object, in order to implement a "Generate_Sequence" function, my Python interpreter tells me that "Observation_Sequence_t" does not exist.
Similar problems appear whenever "deftype" types are required.
Is there something that I have missed regarding my understanding of the library?
Do you have (simple) examples of code where such functions (like "Generate_Sequence", from HMM) get implemented?